key: cord-0848452-e35v0bl2
authors: Yabe, Takahiro; Tsubouchi, Kota; Fujiwara, Naoya; Wada, Takayuki; Sekimoto, Yoshihide; Ukkusuri, Satish V.
title: Non-compulsory measures sufficiently reduced human mobility in Tokyo during the COVID-19 epidemic
date: 2020-10-22
journal: Sci Rep
DOI: 10.1038/s41598-020-75033-5
sha: 6564ac209306036d7c091bb4f5ef29e14f66f5b0
doc_id: 848452
cord_uid: e35v0bl2

While large scale mobility data has become a popular tool to monitor the mobility patterns during the COVID-19 pandemic, the impacts of non-compulsory measures in Tokyo, Japan on human mobility patterns has been under-studied. Here, we analyze the temporal changes in human mobility behavior, social contact rates, and their correlations with the transmissibility of COVID-19, using mobility data collected from more than 200K anonymized mobile phone users in Tokyo. The analysis concludes that by April 15th (1 week into state of emergency), human mobility behavior decreased by around 50%, resulting in a 70% reduction of social contacts in Tokyo, showing the strong relationships with non-compulsory measures. Furthermore, the reduction in data-driven human mobility metrics showed correlation with the decrease in estimated effective reproduction number of COVID-19 in Tokyo. Such empirical insights could inform policy makers on deciding sufficient levels of mobility reduction to contain the disease.

www.nature.com/scientificreports/ by the end of the observation period on April 15th. Panels C and D in Fig. 2 show the complementary cumulative probability distribution plots for both mobility metrics across three different dates, January 20th (before the pandemic), March 30th (before the state of emergency), and April 14th (after the state of emergency), which are all Mondays. The RG values observe a dip at around 20 km, which corresponds to the diameter of the Tokyo Metropolitan area (scale shown in Fig. 1A -C). This implies that while the majority of the individuals move within the metropolitan area, some individuals travel to different regions outside Tokyo. In both mobility metrics, we observe substantial tempering of the tail probability as time passes. This indicates a decrease in relatively long distance travelers ( > 20 km for RG and > 100 km in TTD), reflecting the decrease and closures of airlines and inter-urban railways that travel outside the metropolitan Tokyo area. Figure 2E shows the rates of users staying κ meters within their estimated home locations. This figure suggests that the rate of people staying close to their estimated home locations increased since mid-March and doubled by April 15th regardless of the threshold value κ , reflecting the increase in people working from home and spending time within their neighborhoods. Figure 3 shows the results of the social contact analysis in Tokyo over the observation period (see "Materials and methods"). The panel figures show the temporal transitions in the relative average social contact index. The contact index is calculated by tracking the number of individual users that each user is observed to be co-located in a given temporal window, using the interpolated trajectory data. We have used 100 m as the spatial threshold parameter for detecting potential co-location events in this analysis. The social contact indexes were normalized by the mean peak contact index values observed prior to the pandemic (January 2020). Thus, in Fig. 3 , contact index value of 1 indicates the average amount of observed contacts during peak hours on a typical weekday prior to the pandemic. The temporal periodicity in Fig. 3 A shows the daily and weekly patterns, including the highest point showing the morning rush hour peak, the following peak showing the returning home rush hour, and substantially small amount of contacts (around 0.2) on weekends and holidays, including the New Year breaks. Panels B and C in Fig. 3 show the total daily contact index and the daily peak contact index values, respectively. By comparing the average social contacts across the time horizon, we can observe in Fig. 3 A that the average amount of social contact for each user starts to decrease as COVID-19 became a threat globally in mid February down to 90% of the typical value. The value decreases more rapidly towards the end of February with the government request for remote working, down to 65% of the typical values. The contact index stays at 65% of typical values until the end of March, and as the number of cases in Tokyo starts to increase (Fig. 1) , the contact index further starts to decrease and reaches 30% of typical values by April 15th, almost reaching the government's goal of reducing social contacts by 80% of the typical values 28 . However, regarding the daily peak contacts (morning rush hour peaks on weekdays and lunchtime peaks on weekends and www.nature.com/scientificreports/ holidays), Panel C shows that peak time social contacts have decreased successfully to around 20% of the typical values by April 15th. Moreover, the inequality in mobility changes across different spatial regions within Metropolitan Tokyo were tested, since it has been suggested that inequality in society could aggravate the spread of COVID-19 29 . Panel D of Fig. 3 shows the transition of daily total contacts for users with home locations in the 23 wards in central Tokyo. Although the social contact index of the regions follow similar general patterns, some heterogeneity in the contact decrease can be observed. Panel E shows strong correlation (Pearson's correlation coefficient R = 0.875 * * * ) between the non-normalized amount of contacts on January 15th (before the pandemic) and April 15th, which is after the pandemic. Most of the 23 regions were successful in reducing the social contact index by 80% (shown by gray dashed line), except for wards including Nerima Ward. Panel F further shows the log-log plot between the relative contact index on April 15th and average taxable income per household for each ward in Tokyo. The average taxable income per household was calculated using data provided by the Official Statistics of Japan through the Portal Site 30 . We observe strong negative correlation ( R = −0.696 * * * ) between taxable income per household and contact index, indicating that households in higher income regions were able to reduce the amount of social contacts and risk of COVID-19 transmission more than households in lower income regions. This highlights an inequality across households in terms of the voluntary mobility restrictions. Higher income households are more likely able to reduce their social contacts by reducing mobility, however lower income households may not have this flexibility. This inequality has substantial implications in terms of disease spread and will require a holistic policy planning.

To understand the implications of mobility behavior change and decrease in social contacts on the transmissibility of COVID-19 in Tokyo, we compare the quantified mobility behavior changes with the estimated effective reproduction number R(t) of Tokyo. The R(t) values were estimated using the weekly averaged number of confirmed COVID-19 cases in Tokyo, reported by the metropolitan government (see "Materials and methods"). Figure 4 shows the weekly moving averaged values of (A) social contact index, (B) radius of gyration, and (C) stay-at-home rates plotted along with the dynamic R(t) values. The gray shaded areas show the 95% confidence interval for the estimated R(t). Panels D-F in Fig. 4 show the temporal evolution of the mobility metrics and R(t), where each dot corresponds to daily values. The vertical error bars show the 95% confidence interval of the estimated R(t) values. A non-linear relationship between the mobility metrics and transmissibility are observed, where the R(t) values substantially decrease around a threshold value of 0.33 in the contact index. We observe that R(t) < 1 is achieved www.nature.com/scientificreports/ with reduction in social contacts down to 0.33 on April 3rd. Further reduction in mobility and social contacts are observed after the declaration of the SoE on April 7th, but R(t) had reached a bottom before the declaration. This suggests that restrictions on mobility certainly allow the reduction of number of transmissions up to a point, but further severe restrictions could have less effects on transmissibility reduction. A more targeted and nuanced approach may be needed to determine where and when mobility restrictions should be imposed.

To provide some examples of such nuanced and directly observable metrics of mobility reductions, Panel G shows the relative amount of visits to various types of points of interest (POIs) on April 3rd, which was when sufficient decrease in mobility was observed. We can observe that for all listed types of POIs (Table S2) including business districts (e.g. Tokyo Midtown, Roppongi Hills), shopping areas (e.g. Ginza, Omotesando), and major stations (e.g. Shinjuku, Shibuya), visits were reduced by around 35-40% on average compared to January weekdays (before the pandemic). Haneda Airport (major airport in Tokyo area) on the other hand, had larger decrease in visits (55%) due to travel restrictions which were imposed in early February and March ( Figure S7 ) .

Large-scale mobility data collected from mobile phones provide us with an opportunity to monitor and understand the impacts of NPIs during the COVID-19 pandemic with an unprecedented spatio-temporal granularity and scale. In this study, we utilized such data to quantify the changes in human mobility behavior and social contacts during the COVID-19 spread in Tokyo, Japan, which provides a unique case study where government policies were limited to non-compulsory measures. The analysis concludes that by April 15th (1 week from declaring state of emergency), human mobility behavior have decreased by around 50% both in terms of radius of gyration and total travel distance per user, and also that social contact index in Tokyo had been reduced by more than 70% both in terms of daily total contacts and daily peak time contacts following non-compulsory measures including remote working requests to businesses and closures of public schools. The analysis results also showed that mobility reduction had already taken place before late March (around 40% decrease of social contacts), when the number of imported cases of COVID-19 started to increase. This indicates that even before the actual spread www.nature.com/scientificreports/ of COVID-19 in Tokyo, and even under only non-compulsory measures, a substantial amount of cooperation was provided by the citizens to contain the spread of COVID-19. By comparing the mobility analysis results with the effective reproduction number R(t) estimated from the daily confirmed number of cases in Tokyo, we were further able to show correlation between the mobility reduction and decrease in transmissibility of COVID-19. During late March, R(t) rose above 1 and the number of cases increased, however, mobility reduced following the nation-wide state of emergency, and the R(t) dropped below 1 by the beginning of April. In fact, it was found that reduction of mobility and social contacts beyond certain thresholds (e.g. social contact index of 0.33) had little incremental effects on the decrease in R(t). Looking forward into lifting the state of emergency declaration, these results could inform decision making on how much human mobility and social contact reduction is needed to keep the effective reproduction number of COVID-19 below 1. Figure 4G provides more directly observable measures of mobility reduction at various points of interest which can be used to interpret the sufficient amount of social contact reduction to contain COVID-19.

The presented empirical results should be considered in the light of some limitations. First, the presented analysis was limited to understanding the correlations between mobility and contact reduction and transmissibility. However, microscopic human behavior, such as sanitizing hands more often or higher rates of wearing face masks, could have affected the reduction in transmissibility. Because of such factors other than human mobility, the relationships between mobility reductions and transmissibility could change after lifting the state of emergency. Due to the lack of data on R(t) under no interventions, it is not certain whether the R(t) values reach a similar value of that in European countries and the US. Additional household surveys and interviews need to be conducted to quantify the effects of such behavioral changes. Second, GPS location data collected from mobile phones are prone to spatial errors up to 100 m. This spatial error could have biased the results of the mobility metrics. However, sensitivity analysis suggests that the conclusions are not affected by the spatial thresholds selected in this study ( Figures S3, S4 ). In addition to the 2% sample rate, the granularity of the data was not adequate to compute the actual contacts between individuals (i.e. 2 m distance for all individuals). Moreover, it is known that mobile phone data contain biases in age and income groups 31, 32 . It is important to analyze the mobility of the more vulnerable and elderly population due to the high mortality rate of COVID-19. Although similar conclusions were reached when different spatial thresholds were used via sensitivity analysis, additional validation using small scale but precise data could further clarify the approximate social contact measure computed in this study using mobile phone data. Third, in Tokyo, data on the daily number of new onset cases were not available. Thus, we estimated the onset timings from the daily number of confirmed cases, by shifting the data back 7 days, following reports from Huang et al. 33 . Starnini et al. 34 has shown that the estimation accuracy of R(t) could suffer by using the daily number of confirmed cases. However, this effect is more substantial in the early dynamics of the epidemic, and less towards later periods where our analysis is focused on.

Future research steps include extending the analysis to other cities in Japan such as Sapporo and Osaka, where different patterns of COVID-19 spread have been observed ( Figure S1 ). Cross-comparative analysis including multiple regions could yield novel and more generalizable insights on the relationships between mobility pattern changes and COVID-19 transmissibility. In addition to the spatial dimension, extending the time period of analysis after April 15th and analyzing the dynamics after the lifting of the interventions could strengthen our understanding between social contact decrease and transmissibility reduction. Although this study used aggregated measures to quantify the behavioral changes (e.g. average social contacts), more microscopic analysis of mobility changes could be conducted with additional data.

Mobile phone location data. Location data of smartphones were collected by Yahoo Japan Corporation through the disaster alert app in order to send relevant notifications and information to the users. The users in this study have accepted to provide their location information. The data are anonymized so that individuals cannot be specified, and personal information such as gender, age and occupation are unknown. Each GPS record consists of a user's unique ID (random character string), timestamp, longitude, and latitude. The data acquisition frequency of GPS locations changes according to the movement speed of the user to minimize the burden on the user's smartphone battery. If it is determined that the user is staying in a certain place for a long time, data is acquired at a relatively low frequency, and if it is determined that the user is moving, the data is acquired more frequently. The data has a sample rate of approximately 2% of the population, and past studies suggest that this sample rate is enough to understand the macroscopic urban dynamics, even during an emergency 35 . We selected a panel of users who were active each day in Tokyo metropolitan area before, during and after the COVID-19 pandemic. This leads to a sample of about 200k users, with approximately 50 data points per user each day ( Figure S2 ). Home locations of each individual was detected by applying mean-shift clustering to the nighttime staypoints (observed between 8 p.m. and 6 a.m.), weighted by the duration of stays in each location 36 .

Socio-economic data. Socio-economic data of 23 wards in Tokyo Metropolis were obtained from the Portal Site of Official Statistics of Japan 30 . The number of households and total taxable income collected from the residents were available through the Portal Site. The taxable income per household values used in the analysis in Fig. 3F were calculated by dividing the total taxable income by the number of households in each region (ward) ( Figure S5 , Table S1 ).

The effective reproduction number R(t) is commonly used to quantify transmissibility of an infectious disease, which is defined as the average number of secondary cases generated by a single infectious case 37 . Effects of NPIs such as social distancing can be measured by the effective reproduction number. In order to estimate R(t) of COVID-19 in Tokyo, we employed time series data of daily www.nature.com/scientificreports/ confirmed COVID-19 cases in Tokyo reported by Tokyo metropolitan government 25 . Since the data of the onset cases are unavailable, we used the confirmed date for the estimation. The original daily data were smoothed by taking the centered 7-day moving average. Estimation of R(t) is conducted based on the code implemented by Jung et al., which is available on the Github repository 38 . The code estimates R(t) using the following steps. Since our data on the number of infected cases is based on the laboratory confirmation, there is a time lag associated with the date of infection. The date of the onset of symptoms is estimated by shifting the confirmed daily number of new cases backwards in time for 7 days to account for the diagnosis delay, based on the report by Huang et al. 33 that the median number of days from onset to first hospital admission was 7.0 days. Furthermore, to estimate the daily number of infections, the daily number of onset cases was back-projected using a lognormal density function with mean of 5.6 days and standard deviation of 2.8 days 39 . R(t) was estimated via the renewal equation 40 , which uses the time series of the number of new infected cases in Tokyo on day t, C(t). The expected number of infected cases E[C(t)] are modeled as:

where E[x] denotes the expectation of random variable x, and T is the latest day of observation. The Weibull distribution with shape factor 2.305 and the scale factor 5.452 was assumed to be the probability density function of the generation interval g(τ ) 41 . Moreover, the cumulative Weibull distribution F(t) with shape factor 1.741 and scale factor 8.573, based on the maximum likelihood estimation using data on the reporting date and the date of illness onset collected by the Ministry of Health, Labor and Welfare of Japan 38 , was introduced to account for the right-truncated reporting delay from illness onset to laboratory confirmation. It was assumed that C(t) follows a Poisson distribution, and R(t) was estimated by maximizing the following likelihood function:

The 95% confidence interval of each estimate was also derived using the likelihood function.

vidual human mobility patterns 27, 32 . In this study, we use three key indexes: Radius of Gyration ( R g ), total travel distance (TTD), and stay-at-home rates parameterized by a spatial threshold κ ( SAH κ ). Given a sequence of GPS observation points P i = {p 1 i , p 2 i , . . . , p N i } in a single day of an individual user i, the radius of gyration is calculated using the following equation.

where p i denotes the center of mass of the GPS observation points of user i. The total travel distance of an individual user is the sum of Euclidean distances between all subsequent pairs of GPS observation points on a given day. The stay-at-home rate, parameterized by a spatial threshold κ , is the rate of individual users who had stayed the entire day within a distance κ from the estimated home location.

Quantifying social contact index from trajectory data. To overcome data sparsity, spatio-temporal interpolation of the GPS location observations were performed. Because the GPS data are collected less frequently when movement is detected, we assume that the individual users are static while there are no observations. Using the interpolated individual trajectory data produced from mobile phones, the social contact indexes were computed. The social contact indexes shown in Fig. 3 were computed for 30 min intervals. First, for each time interval [t, t + dt) , where dt = 30 min, users who were not within 100 m from their estimated home locations were detected as "staying out". We denote this set of individual users as N out t . For user i staying out ( i ∈ N out t ), we compute the number of other "staying out" users who are within 100 m from user i, and use that count c i,t as a proxy for social contacts. The social contact index is calculated as the total social contacts for all users staying out, divided by the total number of users including those staying at their homes. Thus, mean social contact value is computed as C t = i∈N out t c i,t /N, where N is the total number of users observed on that day. The social contact index is the relative value of mean social contacts with respect to typical mobility patterns, observed before the COVID-19 pandemic. The typical mean social contact value is computed by taking the average of daily peak social contact values observed on weekdays in January 2020. Thus, the social contact index (SCI) of 1 corresponds to the same amount of social contacts as the daily peak times on weekdays before the COVID-19 pandemic.

Received: 9 July 2020; Accepted: 6 October 2020 

How will country-based mitigation measures influence the course of the COVID-19 epidemic?

Estimating the number of infections and the impact of non-pharmaceutical interventions on COVID-19 in European countries

Mobile phone data for informing public health actions across the COVID-19 pandemic life cycle

Using mobile phone data to predict the spatial spread of cholera

Mobile phone data highlights the role of mass gatherings in the spreading of cholera outbreaks

On the use of human mobility proxies for modeling epidemics

Quantifying the impact of human mobility on malaria

asses sing-chang es-in-commu ting-and-indiv idual -mobil ity-in-major -metro polit an-areas -in-the-unite d-state s-durin g-the-covid -19-outbr eak

The effect of human mobility and control measures on the COVID-19 epidemic in China

Effect of non-pharmaceutical interventions for containing the COVID-19 outbreak in China

Covid-19 outbreak response: A first assessment of mobility changes in Italy following national lockdown

Mobile phone data analytics against the COVID-19 epidemics in italy: Flow diversity and local job markets during the national lockdown

Impacts of state-level policies on social distancing in the United States using aggregated mobility data during the COVID-19 pandemic

Mapping county-level mobility pattern changes in the united states in response to COVID-19

Effects of the COVID-19 pandemic on population mobility under mild policies: Causal evidence from Sweden

Analysis of human mobility in the UK during the COVID-19 pandemic

The relationship between human mobility and viral transmissibility during the COVID-19 epidemics in Italy

COVID-19 lockdown induces structural changes in mobility networks-Implication for mitigating disease dynamics

Controling the spread of COVID-19 using human mobility big data

Analysis of changes in human mobility due to spread of novel coronavirus

Mobile spatial statistics population map

An interactive web-based dashboard to track COVID-19 in real time

Interactive web-based graphs of novel coronavirus COVID-19 cases and deaths per population by country

Open pflow: Creation and evaluation of an open dataset for typical people mass movement in urban areas

COVID-19 information website

Understanding individual human mobility patterns

The Japan Times. Japan's coronavirus total surpasses 10,000 as Tokyo adds 181 cases

Why inequality could spread COVID-19

On the relationship between socio-economic factors and cell phone usage

A survey of results on mobile phone datasets analysis

Clinical features of patients infected with 2019 novel coronavirus in Wuhan

Impact of the accuracy of case-based surveillance data on the estimation of timevarying reproduction numbers

Understanding post-disaster population recovery patterns

Using GPS to learn significant locations and predict movement across multiple users

Handbook of Infectious Disease Data Analysis

Real-time estimation of the effective reproduction number of COVID-19 in Japan

Incubation period and other epidemiological characteristics of 2019 novel coronavirus infections with right truncation: A statistical analysis of publicly available case data

Mathematical Tools for Understanding Infectious Disease Dynamics

Serial interval of novel coronavirus (covid-19) infections

/scientificreports/ by JSPS KAKENHI Grant Number JP18K11462. T.W. was supported by the Osaka City University Strategic Research Grant 2020 for basic researches Kakenhi 20H03932

All authors designed research; T.Y., K.T., N.F., and S.V.U. performed analysis; T.Y., K.T., and N.F. analyzed data; All authors wrote the paper.

The authors declare no competing interests.

Supplementary information is available for this paper at https ://doi.org/10.1038/s4159 8-020-75033 -5.Correspondence and requests for materials should be addressed to S.V.U.Reprints and permissions information is available at www.nature.com/reprints.Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.